A simplified identification and performance prediction method for uplink interference in wireless systems

Abstract

The present invention discloses a method for simplified identification and performance prediction of uplink interference in a wireless system, comprising the following steps: step 1, determining the interference model required for simplified identification and performance prediction of interference; step 2, collecting user data generated when the wireless system is running; step 3, simplified interference identification; step 4, wireless performance prediction. The advantages of the present invention are that simplified interference identification is proposed, with signal user indication vector, interference user indication vector and power offset vector as features, UL-SINR measurement as labels, and supervised machine learning algorithm is used for training to obtain link-level interference relations, and a complete and accurate uplink interference relationship modeling scheme is provided without the need for additional physical equipment and user complete measurement reports, thereby realizing low-complexity, fast, accurate and complete link-level interference identification.

Description

Technical Field

[0001] The invention belongs to the technical fields of wireless communication, multi-user interference, interference prediction and nonlinear regression, and in particular relates to a method for simplified identification and performance prediction of uplink interference in a wireless system. Background Art

[0002] At present, with the rapid growth of network capacity demand, today's wireless networks adopt the method of increasing the deployment density of network equipment to meet the demand for network capacity. Ultra-dense Network (UDN), one of the key technologies of 5G, effectively improves network capacity by reducing the transmission distance between base stations and end users through more densely deployed base stations, and thus improves spectrum efficiency. However, the dense deployment of small base stations will lead to serious co-channel interference (CCI), which brings new challenges to wireless communication networks. With the centralized processing characteristics of the central unit (CU) of the 5G network architecture, machine learning algorithms are applied to mine the interference information hidden in the massive wireless resource allocation data and wireless measurement data generated during network operation, which is not only more accurate and comprehensive, but also more suitable for actual wireless networks. Similarly, the more accurate and information-rich interference model obtained by the new interference modeling scheme also leaves more operating space for resource allocation schemes. Existing wireless communication networks have made high-intensity use of spectrum resources and generally show interference-limited characteristics. Interference has become a core issue in wireless networks and directly determines network performance. There are two existing methods for obtaining an interference matrix: one is to establish an interference matrix based on frequency scanning data, and the other is to establish an interference matrix based on a measurement report message of a mobile phone.

[0003] The frequency domain information in the frequency sweep data is complete and contains longitude and latitude information, which can accurately reflect the interference situation at the sampling point. The interference matrix generated based on the frequency sweep data cannot reflect the interference situation at unknown locations, especially under dense network conditions, where small changes in position can bring about large changes in interference, resulting in excessively high measurement costs. The mobile phone measurement report contains the user's actual interference situation, but the interference information only contains a few neighboring cells with strong signals. Therefore, the interference information is incomplete and the established interference matrix has certain errors. In a dense network, the interference situation is even worse and more complex. When in a dense network with a large number of users, the agility and accuracy of the interference matrix establishment have higher requirements, but the current existing methods are non-real-time, inefficient, coarse-grained, and the rough prediction accuracy cannot adapt to the current network situation.

[0004] In the existing cellular network, the wireless resource allocation function is completed by the base station, and each cell basically manages and allocates wireless resources independently. In order to deal with inter-cell interference, the existing network makes up for it to a certain extent through negotiation and signaling interaction between network units and supplemented by enhanced technology, such as using the X2 interface interaction information between base stations to solve the inter-cell interference problem through Inter Cell Interference Coordination (ICIC) or enhanced ICIC (eICIC) technology; or using Coordinated Multiple Points (CoMP) technology to coordinate interference between different base stations, or avoid interference, or convert interference into useful signals, to provide users with higher rates, thereby improving network utilization.

[0005] ICIC and eICIC technologies rely heavily on signaling exchanges, so the interference information they can transmit is extremely limited, resulting in poor granularity of interference information transmission; and signaling transmission takes time, which seriously affects timeliness. At the same time, a large number of adjacent cells in UDN will cause considerable signaling exchange overhead, affecting network performance. CoMP technology requires a large number of channel measurements, consumes a large number of pilot resources, and consumes a large amount of computing resources to process and calculate signals, so this is not a suitable solution either.

[0006] In addition, the interference calculation method based on the wireless propagation loss model and the terminal geographic location information is only theoretically feasible and not very feasible in actual network deployment for the following reasons:

[0007] 1. In actual wireless systems, it is difficult to obtain the geographical location information of the terminal in real time;

[0008] 2. The propagation path loss model is only applicable to simulation evaluation and network planning scenarios. Obviously, the general model cannot well reflect the radio wave propagation conditions of the application environment. It is very common for the model calculation value to differ from the measured value by 5 to 10 dB, and the prediction accuracy is extremely poor.

[0009] There are some intelligent interference modeling solutions in the existing technology. They use the large amount of data generated when the network is running to obtain the inter-link interference without the need for additional hardware or pilot resources. For example, the existing technology [1] (Resource Allocation for Ultradense Networks with

[0010] The neural network algorithm used in the prior art [2] (Neural Network Based Conflict-Graph Construction Approach for Ultra-Dense Networks. In 2018 IEEE Globecom Workshops, GC Wkshps 2018-Proceedings (pp. 1–6). IEEE) can accurately obtain the interference relationship and strength between users and construct the interference matrix, but the model is not interpretable and the calculation takes a long time. Prior art [3]

[0011] ("Regression-Based Uplink Interference Identification and SINRPrediction for 5G Ultra-Dense Network,"ICC 2020-2020IEEE InternationalConference on Communications(ICC),2020,pp.1-6,) uses a nonlinear regression method combined with wireless knowledge to model interference relationships, making the model interpretable and improving performance. The existing technology [4] ("AXGBoost Based Wireless Interference Relation Mining and PerformancePrediction Method,"2021IEEE 94th Vehicular Technology Conference(VTC2021-Fall),2021,pp.1-7,doi:10.1109 / VTC2021-Fall52928.2021.9625052.) uses the XGBoost method combined with wireless knowledge to model interference.

[0012] The interference identification of the prior art [1]-the prior art [4] aims to obtain the noise to signal ratio (NSR) and the interference to signal ratio (ISR) between links. The different ISRs of different users associated with the same base station are due to the different NSRs of the signal links, rather than the different received powers of the interference links. Directly mining the ISR complicates the interference modeling. In the solutions of the prior art [1]-the prior art [4], the resource scheduling data of user m is only used for the interference relationship training of user m itself, and cannot be reused with other users belonging to the same base station, which has poor scalability for multi-user scenarios in reality. Summary of the invention

[0013] The object of the present invention is to provide a method for simplifying uplink interference identification and wireless performance prediction of a wireless system, which can overcome the defects of the above-mentioned prior art. The method comprises the following steps:

[0014] Step 1: Determine the interference model required for interference reduction identification and performance prediction, that is, the functional relationship between the link-level interference relationship and the uplink signal to interference plus noise ratio (UL-SINR):

[0015] The interference model is built around the link-level interference relationship, which is defined around the base station and describes the link strength from all users to the base station, i.e., the set of SNRs of all signal links and the set of INRs of all interference links received at the base station under open-loop power control. For base station j, the mathematical expression of the link-level interference relationship is: S j is the set of signal users belonging to base station j, N' j is the set of interfering users of base station j. The following is the derivation process of the interference model:

[0016] In the uplink direction of the mobile communication network, the uplink signal transmitted by user m arrives at the home base station j. m The SNR is shown in the following formula (1):

[0017]

[0018] in, Signal user m to home base station j m The channel gain, P m is the transmission power of user m, which is assigned to base station j by user m m The power control or power allocation strategy determines that the interference user n that occupies the same wireless resources as the signal user m arrives at the base station j. mThe interference signal power is shown in the following formula (2):

[0019]

[0020] in, The home base station j of the interference user n to the signal user m m The channel gain, P n is the transmission power of user n, which is assigned to base station j by user n n Power control or power allocation strategy decision;

[0021] Considering the open loop power control (OLPC) strategy, user m is assigned to base station j. m The UL-SINR of the end is shown in the following formula (3):

[0022]

[0023] in, is the signal user m at the home base station j m The SNR of the end, The interference user n is at the home base station j of the signal user m m INR at the end, is base station j m The noise term at includes the noise power and the signal power outside the wireless system, ω n Is to interfere with user instructions, ω n =1 means that user n interferes with the signal user in one transmission, N' m is the set of all users that interfere with signal user m. According to formula (3) for a single user, we get j The formula is used to obtain the functional relationship between the link-level interference relationship centered on the home base station and the UL-SINR, as shown in the following formula (4):

[0024]

[0025] in, Represents the signal user in one transmission, N' j Denotes the set S of all signal users j The set of users causing the interference;

[0026] The UL-SINR, SNR and INR are logarithmically processed to obtain the functional relationship between the link-level interference relationship in dB and the UL-SINR as shown in the following formula (5):

[0027]

[0028] For wireless systems with both OLPC and dynamic power allocation, the UL-SINR is affected by the dynamic transmit power. The functional relationship between the link-level interference relationship and the uplink signal-to-interference-noise ratio is shown in the following formula (6):

[0029]

[0030] Among them, {Δ m |m∈S j}∪{Δ n |n∈N' j} represents the offset of the user's transmit power relative to the open-loop power control transmit power;

[0031] In wireless systems with orthogonal multiplexing resources, N' j is the set of interfering users I that do not belong to base station j j , in wireless systems with non-orthogonal multiplexing resources N' j Including S j and I j , depending on the scenario, formula (6) can be expressed as formula (7) or formula (8). In a wireless system with orthogonal multiplexing resources, as shown in the following formula (7):

[0032]

[0033] Where θ={θ n ∈{0,1}|n∈I j} is the out-of-cell interference user indication vector, θ n =1 indicates that the user n outside the cell causes interference to the signal user in one transmission, in a wireless system with non-orthogonal multiplexing resources, as shown in the following formula (8):

[0034]

[0035] Where δ={δ l ∈{0,1}|l∈S j} represents the interference user indication vector in the cell, δ l =1 means that user l in the cell causes interference to the signal user in one transmission;

[0036] Step 2: Collect user data generated during the operation of the wireless system, including resource allocation vectors, power allocation vectors, measured UL-SINR and transmission time interval (TTI) number, and organize the data set around the home base station:

[0037] Step 2.1, obtain the resource allocation vector α on any resource block in the wireless system = {α m |m∈U} and the power allocation vector Δ={Δm |m∈U}, U is the set of all users in the wireless system, α m =1 means user m uses the resource block, Δ m =k represents that the transmit power of user m is offset by kdB relative to the open-loop power control transmit power;

[0038] Step 2.2, according to the resource allocation vector, for all base stations that are scheduled, that is, satisfy Any base station j, record the signal user indication vector and the interference user indication vector ω = {α n |n∈N' j} and the UL-SINR measured by base station j, and one piece of data is added to the resource allocation data set of all scheduled base stations, including the TTI number, the signal user indication vector, the interference user indication vector and the measured UL-SINR;

[0039] Step 2.3, based on the power allocation vector, record the power offset vectors {Δ m |m∈S j}∪{Δ n |n∈N' j}, add one piece of data to the power allocation data set of all scheduled base stations, including the TTI number and the power offset vector;

[0040] Step 3: Streamlining interference identification: Based on the theoretical basis of the interference model in step 1, a supervised machine learning algorithm is used to mine the base station data set obtained in step 2, and the link-level interference relationship is obtained according to the model parameters;

[0041] The training is performed with the home base station as the center, that is, for any base station, the base station data set constructed in step 2 is used to obtain the link-level interference relationship from all users to the base station. During the training, a supervised machine learning method is used to measure the signal-to-interference-noise ratio as the label data, and the signal user indication vector, the interference user indication vector and the power offset vector are used as the feature data. After the training is completed, a set of model parameters is obtained;

[0042] For algorithms whose model parameters can be explained, such as nonlinear regression algorithms, training is performed based on the interference model in step 1, and the obtained parameters are the defined link-level interference relationships;

[0043] For algorithms whose model parameters cannot be explained, such as neural network algorithms and XGBoost, the parameters of the neural network algorithm are neural network information such as activation function and connection weights, and the parameters of the XGBoost algorithm are regression tree information. The model parameters fit the interference model as a whole, but there is no clear physical meaning. The physical meaning of the output of the algorithm model after training is the predicted UL-SINR. The power offset vector of all zeros and the interference user indication vector of all zeros are input to the algorithm model. The physical meaning of the output is the SNR of the signal link under open-loop power control power. The power offset vector of all zeros and the interference user indication vector with only one interference user indication of 1 and the rest of them are 0 are input to the algorithm model. The physical meaning of the output is the SIR of the signal link and the interference link under open-loop power control power. The INR of the interference link under open-loop power control power is obtained by combining the SNR, which is the defined link-level interference relationship.

[0044] Step 4: wireless performance prediction: under any resource allocation and power allocation conditions, the UL-SINR prediction result is calculated using the link-level interference relationship, or the UL-SINR prediction result is obtained according to the machine learning algorithm model;

[0045] Wireless performance forecasting uses two forecasting methods:

[0046] Prediction method 1: both the algorithm that can be explained by the model parameters and the algorithm that cannot be explained by the model parameters are based on the link-level interference relationship obtained in step 3. The interference user indication vector to be predicted ω={ω n ∈{0,1}|n∈N' j} and the power offset vector Δ to be predicted m ∪{Δ n |n∈N' j} Calculate the prediction for user m As shown in the following formula (9):

[0047]

[0048] Prediction method 2: For algorithms whose model parameters cannot be explained, such as neural network algorithms and XGBoost algorithms, the signal user indication vector to be predicted, the interference user indication vector to be predicted, and the power offset vector to be predicted are directly input into the model, and the model outputs the predicted UL-SINR.

[0049] The method of the present invention has the following beneficial effects:

[0050] 1. The method of the present invention proposes simplified interference identification, using signal user indication vector, interference user indication vector and power offset vector as features, measuring UL-SINR as labels, using any supervised machine learning algorithm for training, and obtaining link-level interference relations. Without the need for additional physical equipment and user complete measurement reports, a complete and accurate uplink interference relationship modeling solution is provided, thereby achieving low-complexity, fast, accurate and complete link-level interference identification;

[0051] 2. A wireless system uplink performance prediction method proposed by the method of the present invention can predict the user UL-SINR under any resource allocation and power allocation conditions;

[0052] 3. The data organization and training of interference reduction identification proposed by the method of the present invention are centered on the home base station, that is, the same user data of the home base station can be used to train a set of link-level interference relations, which can make more effective use of data, while the traditional interference identification algorithm is user-centric, that is, the data of a user is only used to train the interference relationship between the user's own signal link and the interference link;

[0053] 4. The simplified interference identification proposed by the method of the present invention improves the interference identification from the link pair granularity to the link granularity, and reduces the complexity of the problem by q times, where q is the average number of users in each cell, also known as the user density. This improves the accuracy performance of the algorithm and reduces the time consumption of data collection and model training, and can ensure timeliness under high user density;

[0054] 5. The method of the present invention is compatible with different power control technologies, including open-loop power control and dynamic power allocation;

[0055] 6. The method of the present invention is compatible with different access technologies, including access methods of orthogonal resource multiplexing and access methods of non-orthogonal multiplexing;

[0056] 7. The method of the present invention proposes a logarithmic model construction method, which expresses the signal-to-interference-plus-noise ratio SINR, the signal-to-noise ratio SNR and the interference-plus-noise ratio INR in dB, so that no additional non-negative constraints need to be added when solving;

[0057] 8. The interference identification proposed by the method of the present invention has a very broad application prospect. By simplifying the interference identification, the link strength information from all users to all base stations can be obtained. With each base station as an anchor point, the wireless location of the user can be determined. For service flow labeling, a mapping database of service flow and wireless location can be established, which is the basis for research on flow prediction and user behavior analysis. Direct application of the interference model can improve the performance of existing ICIC and eICIC technologies. The results of interference identification can assist in the prediction of wireless performance, obtain the predicted values ​​of physical quantities such as SINR, rate, and energy efficiency, and then assist the intelligent controller in making decisions.

[0058] 9. By applying the method of the present invention, interference identification can quickly and accurately obtain a complete interference model, and can provide support for intelligent management of wireless networks at different time scales, including non-real-time RAN Intelligent Controller and near-real-time RAN Intelligent Controller. By applying the method of the present invention to perform wireless performance prediction, data can be collected in parallel to train the model. After the training is completed, the prediction model used for online prediction can be hot-updated, and iterative updates can be made to achieve rapid response and tracking of network conditions.

[0059] 10. Compared with the interference calculation method based on the wireless propagation loss model and the terminal geographic location information, the method of the present invention does not require the geographic location information of the user terminal. At the same time, the established interference model is more targeted and closer to the wireless network environment to be applied than the general wireless propagation loss model. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a schematic diagram of link-level interference of the method of the present invention;

[0061] Figure 2 It is a schematic diagram comparing the interference source identification performance of each scheme under different data scales of the method of the present invention;

[0062] Figure 3 Schematic diagram of UL-S I NR prediction error of each scheme under different data set scales according to the method of the present invention;

[0063] Figure 4 It is a schematic diagram of the average time consumed by each user in each scheme to obtain an interference model under different training data set sizes according to the method of the present invention;

[0064] Figure 5 It is a schematic diagram of the average training time required for base stations under certain performance indicators for each interference modeling and prediction scheme of the method of the present invention;

[0065] Figure 6 1 is a schematic diagram of interference source identification performance of the method of the present invention with different training data amounts of r when q=8;

[0066] Figure 7 1 is a schematic diagram of UL-SINR prediction errors of various schemes under different data set sizes when q=8 according to the method of the present invention;

[0067] Figure 8 It is a schematic diagram of the training time consumption of each scheme under different data set scales when q=8 according to the method of the present invention;

[0068] Fig. 9 It is a schematic diagram comparing the average training time to achieve the specified performance when q=8 according to the method of the present invention. DETAILED DESCRIPTION

[0069] The specific implementation of the method of the present invention is described in detail below with reference to the accompanying drawings.

[0070] Step 1: Determine the interference model required for interference reduction identification and performance prediction, that is, the functional relationship between link-level interference relationship and UL-SINR:

[0071] The interference model is built around the link-level interference relationship, which is defined around the base station and describes the link strength from all users to the base station, i.e., the set of SNRs of all signal links and the set of INRs of all interference links received at the base station under open-loop power control. For base station j, the mathematical expression of the link-level interference relationship is: S j is the set of signal users belonging to base station j, N' j is the set of interfering users of base station j; the following is the derivation process of the interference model:

[0072] In the uplink direction of the mobile communication network, the uplink signal transmitted by user m arrives at the home base station j. m The SNR is shown in the following formula (1):

[0073]

[0074] in, Signal user m to home base station j m The channel gain, P m is the transmission power of user m, which is assigned to base station j by user m m The power control or power allocation strategy determines that the interference user n that occupies the same wireless resources as the signal user m arrives at the base station j. m The interference signal power is shown in the following formula (2):

[0075]

[0076] in, The home base station j of the interference user n to the signal user m m The channel gain, P n is the transmission power of user n, which is assigned to base station j by user n n Power control or power allocation strategy decision;

[0077] Considering the open loop power control (OLPC) strategy, user m is assigned to base station j. m The UL-SINR of the end is shown in the following formula (3):

[0078]

[0079] in, is the signal user m at the home base station j m The SNR of the end, The interference user n is at the home base station j of the signal user m m INR at the end, is base station j m The noise term at includes the noise power and the signal power outside the wireless system, ω n Is to interfere with user instructions, ω n =1 means that user n interferes with the signal user in one transmission, N' m is the set of all users that interfere with signal user m. According to formula (3) for a single user, we get j The formula is used to obtain the functional relationship between the link-level interference relationship centered on the home base station and the UL-SINR, as shown in the following formula (4):

[0080]

[0081] in, Represents the signal user in one transmission, N' j Denotes the set S of all signal users j The set of users causing the interference;

[0082] The UL-SINR, SNR and INR are logarithmically processed to obtain the functional relationship between the link-level interference relationship in dB and the UL-SINR as shown in the following formula (5):

[0083]

[0084] For wireless systems with both OLPC and dynamic power allocation, the UL-SINR is affected by the dynamic transmit power. The functional relationship between the link-level interference relationship and the uplink signal-to-interference-noise ratio is shown in the following formula (6):

[0085]

[0086] Among them, {Δ m |m∈S j}∪{Δ n |n∈N' j} represents the offset of the user's transmit power relative to the open-loop power control transmit power;

[0087] In wireless systems with orthogonal multiplexing resources, N' j is the set of interfering users I that do not belong to base station j j , in wireless systems with non-orthogonal multiplexing resources N' j Including S j and I j , depending on the scenario, formula (6) can be expressed as formula (7) or formula (8). In a wireless system with orthogonal multiplexing resources, as shown in the following formula (7):

[0088]

[0089] Where θ={θ n ∈{0,1}|n∈I j} is the out-of-cell interference user indication vector, θ n =1 indicates that the user n outside the cell causes interference to the signal user in one transmission, in a wireless system with non-orthogonal multiplexing resources, as shown in the following formula (8):

[0090]

[0091] Where δ={δ l ∈{0,1}|l∈S j} represents the interference user indication vector in the cell, δ l =1 means that user l in the cell causes interference to the signal user in one transmission;

[0092] Define the link-level interference relationship centered on the home base station, like Figure 1As shown, in the wireless system, the solid arrow represents the signal link between the user and the belonging base station, and the dotted arrow represents the interference link between the user and the non-belonging base station. There are four signal links in the wireless system, and any signal link has two interference links, and the link-level interference is 2x4=8; there are two base stations in the wireless system, and any base station has two interference links, and the link-level interference is only 2x2=4. According to the specific wireless system, select the appropriate interference model in formulas (6)-(8);

[0093] Step 2: Collect user data generated during the operation of the wireless system, including resource allocation vectors, power allocation vectors, measured UL-SINR and transmission time interval (TTI) number, and organize the data set around the home base station:

[0094] Step 2.1, obtain the resource allocation vector α on any resource block in the wireless system = {α m |m∈U} and the power allocation vector Δ={Δ m |m∈U}, U is the set of all users in the wireless system, α m =1 means user m uses the resource block, Δ m = k represents that the transmit power of user m is shifted by k dB relative to the open-loop power control transmit power;

[0095] Step 2.2, according to the resource allocation vector, for all base stations that are scheduled, that is, satisfy For any base station j, record the signal user indication vector and the interference user indication vector ω = {α n |n∈N' j} and the UL-SINR measured by base station j, and one piece of data is added to the resource allocation data set of all scheduled base stations, including the TTI number, the signal user indication vector, the interference user indication vector and the measured UL-SINR;

[0096] Step 2.3: Based on the power allocation vector, record the power offset vectors {Δ m |m∈S j}∪{Δ n |n∈N' j}, add one piece of data to the power allocation data set of all scheduled base stations, including the TTI number and the power offset vector;

[0097] Step 3: Simplify interference identification. Based on the theoretical basis of the interference model in step 1, use a supervised machine learning algorithm to mine the base station data set obtained in step 2 and obtain the link-level interference relationship according to the model parameters.

[0098] The training is performed with the home base station as the center, that is, for any base station, the base station data set constructed in step 2 is used to obtain the link-level interference relationship from all users to the base station. During the training, a supervised machine learning method is used to measure the signal-to-interference-noise ratio as the label data, and the signal user indication vector, the interference user indication vector and the power offset vector are used as the feature data. After the training is completed, a set of model parameters is obtained;

[0099] For algorithms whose model parameters can be explained, such as nonlinear regression algorithms, training is performed based on the interference model in step 1, and the obtained parameters are the defined link-level interference relationships;

[0100] For algorithms whose model parameters cannot be explained, such as neural network algorithms and XGBoost, the parameters of the neural network algorithm are neural network information such as activation function and connection weights, and the parameters of the XGBoost algorithm are regression tree information. The model parameters fit the interference model as a whole, but there is no clear physical meaning. The physical meaning of the output of the algorithm model after training is the predicted UL-SINR. The power offset vector of all zeros and the interference user indication vector of all zeros are input to the algorithm model. The physical meaning of the output is the SNR of the signal link under open-loop power control power. The power offset vector of all zeros and the interference user indication vector with only one interference user indication of 1 and the rest of them are 0 are input to the algorithm model. The physical meaning of the output is the SIR of the signal link and the interference link under open-loop power control power. The INR of the interference link under open-loop power control power is obtained by combining the SNR, which is the defined link-level interference relationship.

[0101] Step 4: wireless performance prediction: under any resource allocation and power allocation conditions, the UL-SINR prediction result is calculated using the link-level interference relationship, or the UL-SINR prediction result is obtained according to the machine learning algorithm model;

[0102] Wireless performance forecasting uses two forecasting methods:

[0103] Prediction method 1: both the algorithm that can be explained by the model parameters and the algorithm that cannot be explained by the model parameters are based on the link-level interference relationship obtained in step 3. Interference user indication vector to be predicted ω={ω n ∈{0,1}|n∈N' j} and the power offset vector Δ to be predicted m ∪{Δ n |n∈N' j} Calculate the prediction for user m As shown in the following formula (9):

[0104]

[0105] Prediction method 2: For algorithms whose model parameters cannot be explained, such as neural network algorithms and XGBoost algorithms, the signal user indication vector to be predicted, the interference user indication vector to be predicted, and the power offset vector to be predicted are directly input into the model, and the model outputs the predicted UL-SINR.

[0106] The simulation scenario in the specific implementation of the present invention is described as follows: According to the 3GPP Dual Strip model, in a small wireless service hotspot area, in order to meet the huge demand for services and throughput, the operator will deploy a large number of network devices to form a UDN. There are two subspace bands on each side of the 20m wide corridor, and each subspace band consists of N r Each subspace is composed of 10m side length subspaces, that is, each subspace is 10m×10m in size, each subspace is a microcell, in which a small base station (SBS) is deployed, and each small base station accesses q users, a total of R=4N r Small base stations and N = 4N r For q users, the allocation of resources within the cell is orthogonal, and there is no interference between users belonging to the same small base station.

[0107] The interference modeling and prediction scheme (hereinafter referred to as “simplified identification-nonlinear regression”) proposed by the method of the present invention is evaluated and analyzed for performance, and simulations are performed in low user density and high user density scenarios respectively. The indicators include SINR prediction error, training time, SNR and INR errors.

[0108] Table 1 below shows the parameters used in the simulation of the low user density scenario:

[0109] Table 1

[0110]

[0111]

[0112] like Figure 2 As shown, by evaluating the performance of interference source identification, it is possible to fully reflect whether the interference model obtained by offline training is accurate. The performance of interference source identification is measured using the root mean square error (RMSE) of each user SIR, as shown in the following formula (17) for user m:

[0113]

[0114] Among them, j m is the user-associated base station, is the predicted value, γ m,n is the ideal SIR value without fast fading, which does not include fast fading and is calculated only by path loss.

[0115] At a data scale of 400,000, the traditional recognition-neural network solution is 0.25dB, the traditional recognition-XGBoost and traditional recognition-nonlinear regression solutions are 0.1dB, and the interference simplified recognition-nonlinear regression is slightly lower, about 0.07dB. At the same time, to achieve the same interference source identification performance, the interference simplified recognition-nonlinear regression can achieve more than 2 times the data volume advantage compared to other algorithms.

[0116] The performance of UL-SINR prediction can reflect the accuracy of the online prediction part of the proposed solution.

[0117] The UL-SINR prediction error performance is measured using the root mean square error RMSE: for user m, as shown in the following formula (18):

[0118]

[0119] is the predicted value of the i-th test data, is the ideal UL-SINR value of the i-th test data It does not include small-scale fading and is calculated only from path loss, inter-user interference, and noise.

[0120] The UL-SINR prediction performance under different training dataset sizes is different:

[0121] like Figure 3 As shown in the figure, it can be seen that the interference simplified identification-nonlinear regression scheme is always the best, which is manifested in the minimum training error under the same data volume or the least data required to achieve a certain training error. Taking the RMSE of 0.5dB as an example, the proposed algorithm only needs 4000 training data per base station, which has a data volume advantage of more than 2 times compared with other algorithms.

[0122] When the offline training time of an interference model of a scheme takes too long, the timeliness of the trained interference model cannot be guaranteed, and the scheme cannot adapt to the rapidly changing characteristics of wireless networks. Therefore, training time is also very important for interference modeling and prediction schemes.

[0123] Figure 4 The average base station time consumption of each prediction scheme to obtain the interference model under different training data set sizes is shown. The nonlinear regression scheme of interference streamlined identification has the lowest training time of all schemes. At a data scale of 40,000, its training time is about 1 / 2 of that of the traditional identification-XGBoost scheme and 1 / 20 of that of the traditional identification-neural network scheme.

[0124] By jointly considering prediction accuracy and training time, the most reasonable size of the training data set can be determined to avoid insufficient prediction accuracy due to a too small size, and to avoid increased time spent on data collection and training due to a too large size, which affects the timeliness of the model.

[0125] Figure 5 The average base station training time required for each interference modeling and prediction scheme under certain performance indicators is shown. Here, the RMSE of UL-SINR is less than 0.5dB as the standard. The average training time of the four schemes is 25, 0.55, 0.34 and 0.19 seconds. Interference simplified identification-nonlinear regression is the best, which has an order of magnitude advantage over traditional identification-neural network, and is slightly better than traditional identification-nonlinear regression and traditional identification-XGBoost.

[0126] In the above comparison, the performance of the interference reduction recognition-nonlinear regression scheme has great advantages over the traditional recognition-neural network scheme, and is slightly better than the traditional recognition-nonlinear regression scheme and the traditional recognition-XGBoost scheme.

[0127] The above simulation analysis is performed in a low user density scenario, where the number of users in each cell is q=2. In the existing network, it is common to have 8-10 user devices under one small base station. As the user density increases, the advantages of the method described in the present invention will become more obvious.

[0128] Table 2 below shows the parameters used in the simulation of the high user density scenario:

[0129] Table 2

[0130]

[0131] like Figure 6 As shown, the data is more fully utilized and the interference reduction recognition performance advantage is more obvious.

[0132] like Figure 7 As shown, taking the prediction error less than 0.5 dB as the standard, the method of the present invention only needs 20,000 items / base station to meet the standard, while the traditional recognition requires 160,000 items / base station. Figure 7 The figure shows that streamlined recognition brings a gain of slightly more than 8 times.

[0133] like Figure 8 As shown, the training time of the traditional recognition nonlinear scheme is 5-8 times that of the interference reduction recognition nonlinear scheme.

[0134] like Fig. 9As shown in the figure, to achieve the same accuracy, interference simplified identification-nonlinear regression has an advantage in training time of more than an order of magnitude. The traditional identification-nonlinear regression solution requires 150s / user training time to achieve an accuracy of 0.5dB when there are 160,000 items / base station, while the simplified identification-nonlinear regression only requires 5.4s / user training time, a difference of about 30 times. Simplified identification can still guarantee timeliness under high user density, while traditional identification is greatly affected by the increase in user density and it is difficult to guarantee timeliness.

[0135] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the scope disclosed in the present invention should be covered by the protection scope of the claims of the present invention.

Claims

1. A method for simplifying identification and performance prediction of uplink interference in a wireless system, characterized in that: The following steps are involved: Step 1: determine the interference model required for interference reduction identification and performance prediction, that is, the functional relationship between the link-level interference relationship and the uplink signal-to-interference-noise ratio; The interference model is built around the link-level interference relationship. The link-level interference relationship is defined around the base station and describes the link strength from all users to the base station, that is, the SNR set of all signal links and the INR set of all interference links received at the base station under open-loop power control. For base station j, the mathematical expression of the link-level interference relationship is S j is the set of signal users belonging to base station j, N' j is the set of interfering users of base station j; For wireless systems with open-loop power control and dynamic power allocation, the functional relationship between link-level interference and uplink signal-to-interference-noise ratio is shown in the following formula: in, is the signal user indicator vector, Indicates that there are m signal users in one transmission, and there is only one signal user in one transmission. ω={ω n ∈{0,1}|n∈N' j } is the interference user indication vector, ω n =1 means user n is an interfering user in a transmission, {Δ m |m∈S j }∪{Δ n |n∈N' j } is a power offset vector, which indicates the offset of the user's transmit power relative to the open-loop power control transmit power. For a wireless system with only open-loop power control, the power offset vector is always 0; In wireless systems with orthogonal multiplexing resources, N' j is the set of interfering users I that do not belong to base station j j , in wireless systems with non-orthogonal multiplexing resources N' j Including S j and I j , In a wireless system with orthogonal multiplexing resources, as shown in the following formula: where θ = {θ n ∈{0,1}|n∈I j } is the out-of-cell interference user indication vector, θ n =1 means that the user n outside the cell causes interference to the signal user in one transmission; In a wireless system with non-orthogonal multiplexing resources, as shown in the following formula: Where δ={δ l ∈{0,1}|l∈S j } represents the interference user indication vector in the cell, δ l =1 means that user l in the cell causes interference to the signal user in one transmission; Step 2, collecting user data generated during the operation of the wireless system, including resource allocation vectors, power allocation vectors, measured UL-SINR and TTI numbers, and organizing the data set around the home base station, including; Step 2.1, obtain the resource allocation vector α on any resource block in the wireless system = {α m |m∈U} and the power allocation vector Δ={Δ m |m∈U}, U is the set of all users in the wireless system, α m =1 means user m uses the resource block, Δ m =k represents that the transmit power of user m is offset by kdB relative to the open-loop power control transmit power; Step 2.2, according to the resource allocation vector, for all base stations that are scheduled, that is, satisfy For any base station j, record the signal user indication vector and the interference user indication vector ω = {α n |n∈N' j } and the UL-SINR measured by base station j; one piece of data is added to the resource allocation data set of all scheduled base stations, including the TTI number, the signal user indication vector, the interference user indication vector and the measured UL-SINR; Step 2.3: Based on the power allocation vector, record the power offset vectors {Δ m |m∈S j }∪{Δ n |n∈N' j }, add one piece of data to the power allocation data set of all scheduled base stations, including the TTI number and the power offset vector; Step 3: Simplify interference identification. Based on the interference model theory of step 1, use a supervised machine learning algorithm to mine the base station data set obtained in step 2 to obtain model parameters, and obtain the link-level interference relationship according to the model parameters. Step 4, wireless performance prediction, under any resource allocation and power allocation conditions, uses the link-level interference relationship and the wireless interference model to calculate the UL-SINR prediction result, or the UL-SINR prediction result is output by the machine learning algorithm model.

2. A method for simplifying identification and performance prediction of uplink interference in a wireless system according to claim 1, characterized in that: In step 3, supervised machine learning is used, and during training, the measured signal to interference and noise ratio is used as label data, and the signal user indication vector, the interference user indication vector and the power offset vector are used as feature data. After the training is completed, a set of model parameters is obtained.

3. The method for simplified identification and performance prediction of uplink interference in a wireless system according to claim 1, characterized in that: In step 3, for the algorithm that can be explained by the model parameters, the parameters obtained through training are the defined link-level interference relations.

4. The method for simplifying identification and performance prediction of uplink interference in a wireless system according to claim 1, characterized in that: In the step 3, for the algorithm whose model parameters cannot be explained, the physical meaning of the output of the interference model after training is the predicted UL-SINR, and the power offset vector of all zeros and the interference user indication vector of all zeros are input to the interference model, and the physical meaning of the output is the SNR of the signal link under the open-loop power control power; the power offset vector of all zeros and the interference user indication vector with only one interference user indication of 1 and the rest of 0 are input to the interference model, and the physical meaning of the output is the SIR of the signal link and the interference link under the open-loop power control power, and the INR of the interference link under the open-loop power control power is obtained by combining the SNR, which is the defined link-level interference relationship.

5. The method for simplified identification and performance prediction of uplink interference in a wireless system according to claim 1, characterized in that: In step 4, the wireless performance prediction adopts two prediction methods: Prediction method 1: both algorithms that can be explained by model parameters and algorithms that cannot be explained by model parameters can be used according to the link-level interference relationship obtained in step 3. Interference user indication vector to be predicted ω={ω n ∈{0,1}|n∈N' j } and the power offset vector Δ to be predicted m ∪{Δ n |n∈N' j } Calculate the predicted UL-SINR of user m As shown in the following formula: Prediction method 2, for algorithms whose model parameters cannot be explained, the signal user indication vector to be predicted, the interference user indication vector to be predicted and the power offset vector to be predicted are input into the model, and the model outputs the predicted UL-SINR.

Images